-
Notifications
You must be signed in to change notification settings - Fork 370
2016 11 14 Open NEST Developer Video Conference
terhorstd edited this page Nov 14, 2016
·
8 revisions
-
Welcome
-
Report from NEST User Workshop 2016
- very successful workshop
- many new contacts between and among users and developers
-
Review of NEST User Mailing List
Mails since last meeting were reviewed and handled. Some additional (meta-)questions came up:
- at some point we need to consider opening a NEST on MAC team, also questions about python on mac came up.
- what is the scope of nest-user? How far do we discuss neuroscience questions about the behavior of specific networks on the software mailing-list?
- No answer on LayoutNetwork, so we take it as a GO and can change behavior in refactoring endeavors
- How many more models to put into the kernel before moving to NESTML? And, at what point do we want to create a separate repository for user models?
-
Update: Work on
GIDCollection
and subnet removal
- See Project 2 and especially at the moment #455
- terhorstd, stinebuu, hakonsbm, heplesser working on it at present
- first major step is conversion of all functions (esp Create, Connect) to use GIDCollections
- currently only primitive GIDCollections (composite being implemented)
- requires adjustment of many tests and most examples
- works partially at SLI level, Python interface being implemented
- When subnets disappear, also related functions such as
GetChildren
,GetLeaves
, probably alsoGetNodes
disappear.- A: no problem
- Main work branch currently heplesser/SubnetFreeNEST
- Discussion: How to proceed towards NEST 2.12 and NEST 3
- Need to identify PRs and Issues to be included in NEST 2.12
- Need issue review session/team (info: new issue labeling scheme coming)
- Need to branch before integrating any subnet-removal-related material
-
Decision: use the github milestones to sort the issues/PRs to NEST 2.12/3.0.
@heplesser, @terhorstd and @jougs will form a task force for doing this.
- Discussion: Removal of
DataConnect
(see #537)- need to rerun the benchmarks with more current code. @hbos is on it.
- Discussion:
LOG( M_ERROR, ...)
- Shouldn't one just throw an exception instead?
- Is this a left-over of old practice?
- See conn_builder.cpp:589 for one of over 50 examples
- A: probably a leftover, we can remove
-
#523 NEST connection infrastructure improvements: Presentation by Alexey Shusharin
- Questions to answer:
- what is the most common use-case for usage of connection
- how important are cases with small number of connections (e.g. spike recorder/generator)
- what is a use-case for target connection ports?
-
Q: if neuron receives inputs through static synapse, all inputs could be placed in one TargetConnector?
- A: Yes, upto e.g. 64 connections of a single synapse model, then new TargetConnector. Other model would be another TargetConnector.
-
Q: How can we deliver spikes efficiently in a linked-list memory layout?
- A: ConnectionHandle provides direct link to the target Connector.
- C: very difficult to predict performance, because locality and cache performance are important. Can only find out by experimenting+benchmarking.
-
Q: how do you handle the sparse table?
- A: The sparse table basically doesn't change, just the contained information is different.
-
C: bottleneck is the delivery and it's very difficult to predict the performance of the proposal due to interactions with ring-buffer entries.
- going to post-synaptic storage would change a lot
- proposal would probably be a good improvement (memory-wise) compared to 4g kernel
- A: @alexeyshusharin wants to implement a prototype, then we can compare performance.
- Questions to answer:
- Review of open Github Pull Request
- -will be handled in issue sorting task force-
- Review of open Github Issues
- -will be handled in issue sorting task force-
NEST Homepage: www.nest-simulator.org
NEST Initiative: www.nest-initiative.org